Summary of Source Matters: Source Dataset Impact on Model Robustness in Medical Imaging, by Dovile Juodelyte et al.
Source Matters: Source Dataset Impact on Model Robustness in Medical Imaging
by Dovile Juodelyte, Yucheng Lu, Amelia Jiménez-Sánchez, Sabrina Bottazzi, Enzo Ferrante, Veronika Cheplygina
First submitted to arxiv on: 7 Mar 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper explores the effectiveness of transfer learning in medical imaging classification algorithms, specifically focusing on whether performance gains stem from improved generalization or shortcut learning. The authors introduce a taxonomy for confounders in medical imaging, MICCAT, and investigate various confounders across two public datasets. They find that ImageNet and RadImageNet achieve similar classification performance, but ImageNet is more prone to overfitting to confounders. The study recommends reexamining model robustness by conducting similar experiments. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper investigates how transfer learning helps medical imaging algorithms work better. It tries to figure out if the improvement comes from being good at recognizing images in general or just memorizing specific features. To do this, it creates a special list of confounders and tests different types on two big datasets. The results show that two popular methods, ImageNet and RadImageNet, are about equally good, but one is more likely to get stuck on familiar patterns. |
Keywords
* Artificial intelligence * Classification * Generalization * Overfitting * Transfer learning
